[ 
https://issues.apache.org/jira/browse/FLINK-28053?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Shengkai Fang updated FLINK-28053:
----------------------------------
    Description: 
There are two kinds of the parallel we need to consider:

1. The parallel among all the operations for the same session. For example, the 
user may submit a SQL to create a table and modify another table's schema in 
parallel.
2. The parallel is also mainly about the Operation itself. It is possible that 
one thread is reading the data from the Operation and another one closes the 
Operation in parallel. 

We may introduce the queue to make these requests execute in sequence. It 
brings the benefit that simplifying the logic in the Operation and 
OperationManager and moving all locks to the handover. But it may cause a 
performance regression. Therefore, it's better if we can start this issue until 
all components finish.





> Introduce queue to execute request in sequence
> ----------------------------------------------
>
>                 Key: FLINK-28053
>                 URL: https://issues.apache.org/jira/browse/FLINK-28053
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Table SQL / Gateway
>    Affects Versions: 1.16.0
>            Reporter: Shengkai Fang
>            Assignee: Shengkai Fang
>            Priority: Major
>
> There are two kinds of the parallel we need to consider:
> 1. The parallel among all the operations for the same session. For example, 
> the user may submit a SQL to create a table and modify another table's schema 
> in parallel.
> 2. The parallel is also mainly about the Operation itself. It is possible 
> that one thread is reading the data from the Operation and another one closes 
> the Operation in parallel. 
> We may introduce the queue to make these requests execute in sequence. It 
> brings the benefit that simplifying the logic in the Operation and 
> OperationManager and moving all locks to the handover. But it may cause a 
> performance regression. Therefore, it's better if we can start this issue 
> until all components finish.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

Reply via email to